Disentangling Prompt Dependence to Evaluate Segmentation Reliability in Gynecological MRI

Segmentation is central to clinical diagnosis and monitoring, yet the reliability of modern foundation models in medical imaging still depends on the availability of precise prompts. The Segment Anything Model (SAM) offers powerful zero-shot capabilities, although it collapses under the weak, generic, and noisy prompts that dominate real clinical workflows. In practice, annotations such as centerline points are coarse and ambiguous, often drifting across neighboring anatomy and misguiding SAM toward inconsistent or incomplete masks. We introduce SPD, a Saliency-Guided Prompt Distillation framework that converts these unreliable cues into robust guidance. SPD first learns data-driven anatomical priors through a lightweight saliency head to obtain confident localization maps. These priors then drive Contextual Prompt Distillation, which validates and enriches noisy prompts using cues from anatomically adjacent slices, producing a consensus prompt set that matches the behavior of expert reasoning. A Pairwise Slice Consistency objective further enforces local anatomical coherence during segmentation. Experiments on four challenging MRI and CT benchmarks demonstrate that SPD consistently outperforms existing SAM adaptations and supervised baselines, delivering large gains in both region-based and boundary-based metrics. SPD provides a practical and principled path toward reliable foundation model deployment in clinical environments where only imperfect prompts are available.

Promptable Foundation Models (FMs), initially introduced for natural image segmentation, have also revolutionized medical image segmentation. The increasing number of models, along with evaluations varying in datasets, metrics, and compared models, makes direct performance comparison between models difficult and complicates the selection of the most suitable model for specific clinical tasks. In our study, 11 promptable FMs are tested using non-iterative 2D and 3D prompting strategies on a private and public dataset focusing on bone and implant segmentation in four anatomical regions (wrist, shoulder, hip and lower leg). The Pareto-optimal models are identified and further analyzed using human prompts collected through a dedicated observer study. Our findings are: 1) The segmentation performance varies a lot between FMs and prompting strategies; 2) The Pareto-optimal models in 2D are SAM and SAM2.1, in 3D nnInteractive and Med-SAM2; 3) Localization accuracy and rater consistency vary with anatomical structures, with higher consistency for simple structures (wrist bones) and lower consistency for complex structures (pelvis, tibia, implants); 4) The segmentation performance drops using human prompts, suggesting that performance reported on "ideal" prompts extracted from reference labels might overestimate the performance in a human-driven setting; 5) All models were sensitive to prompt variations. While two models demonstrated intra-rater robustness, it did not scale to inter-rater settings. We conclude that the selection of the most optimal FM for a human-driven setting remains challenging, with even high-performing FMs being sensitive to variations in human input prompts. Our code base for prompt extraction and model inference is available: https://github.com/CarolineMagg/segmentation-FM-benchmark/

Segmentation is a fundamental vision task underlying numerous downstream applications. Recent promptable segmentation models, such as Segment Anything Model 3 (SAM3), extend segmentation from category-agnostic mask prediction to concept-guided localization conditioned on high-level textual prompts. However, existing benchmarks primarily evaluate mask accuracy or object presence, leaving unclear whether these models faithfully ground the queried concept or instead rely on visually salient but semantically misleading cues. We introduce CAFE: Counterfactual Attribute Factuality Evaluation, a novel benchmark for evaluating concept-faithful segmentation in promptable segmentation models. Our CAFE is built on attribute-level counterfactual manipulation: the target region and ground-truth mask are preserved, while attributes such as surface appearance, context, or material composition are modified to introduce misleading semantic cues. The benchmark contains 2,146 paired test samples, each consisting of a target image, a ground-truth mask, a positive prompt, and a misleading negative prompt. These samples cover three counterfactual categories: Superficial Mimicry (SM), Context Conflict (CC), and Ontological Conflict (OC). We evaluate various model types and sizes on our CAFE. Experiments reveal a systematic gap between localization quality and concept discrimination: models often generate accurate masks even for misleading prompts, suggesting that strong mask prediction does not necessarily imply faithful semantic grounding. Our CAFE provides a controlled benchmark for diagnosing whether promptable segmentation models perform concept-faithful grounding rather than shortcut-driven mask retrieval.